This is a repository for the paper "Evaluating the Robustness of Deep Reinforcement Learning for Autonomous Policies in a Multi-agent Urban Driving Environment"
-
Install the system requirements:
- Ubuntu 18.04+
- Anaconda (latest version)
- cmake (
sudo apt install cmake) - zlib (
sudo apt install zlib1g-dev) - [optional] ffmpeg (
sudo apt install ffmpeg)
-
Setup CARLA (0.9.4):
Run mkdir ~/software && cd ~/software
Download the 0.9.4 release version from: Here
Extract it into ~/software/CARLA_0.9.4
Run echo "export CARLA_SERVER=${HOME}/software/CARLA_0.9.4/CarlaUE4.sh" >> ~/.bashrc
- Install the libraries:
Fork/Clone the repository to your workspace:
Create a new conda env named "Benchmarking" and install the required packages:
conda env create -f conda_env.yml
Activate the environment:
conda activate macad-gym-benchmarking
Run the following commands in sequence for installing rest of the packages to avoid version errors:
pip install -e .
pip install --upgrade pip
pip install -e .
pip install tensorflow==2.1.0
pip install tensorflow-gpu==2.1.0
pip install pip install ray[tune]==0.8.4
pip install pip install ray[rllib]==0.8.4
pip install tf_slim
pip install tensorboardX==2.1
Soon to be updated here along with the paper.